Quality of of synthetic winds by testing of similaryty to measured data by spectral modelling, rainflow count analysis and statistics of increments

Introduction

While wind energy industry growing rapidly and siting of wind turbines onshore as well as offshore is increasing,
many wind engineering model tools have been developed for the assessment of loads on wind turbines due to
varying wind speeds. In order to have proper wind turbine design and performance analysis, it is important to have
an accurate representation of the incoming wind field. To ease the analysis, tools for the generation of synthetic
wind fields have been developed, e.g the widely used NREL TurbSim procedure.
We analyse respective synthetic data sets on one hand in view of the similarity of the spectral characteristics of
measured and synthetic sets. In addition, second order characteristics with direct relevance to load assessment as
given by the statistics of increments are inspected.

Approach

Characteristics of measured data are taken from the library of the server of the US NREL wind technology center. Setscovering 10 minites of of 3D 20HZ wind speed at different heights and respective anxilary data are used here. For these sets the velocity spectra, the PDF's of the wind speed increments and rainflow cycle counts are extracted.
The results are sorted according to classes of height, wind speed and thermal stratification.
The class parameters are use as input forNREL Turbsim procedure. Here various spectral models - most notably the Mann model, and both, the basic data generation scheme and a scheme negociating coherend events are applied.

Main body of abstract

Concerning the spectra, the model of Mann gives overall satisfactory results for reflecting the empirical spectra.
Concerning the statistics of increments, as expected the pdf's of the increments are non Gussian showing an incraesed probability for extreeme increments. The deviations from Gaussian show a variation for the different classes that requires further inspection.
The synthetic data generated without the inclusion of coherend events - as expected - fail to reflect any non Gaussian characteristics.
The data generated including coherend events show pdf's of the increments that are able to quantitatively reflect the non Gussian empirical pdf's. A closer addaptions to the empirical characteristics has to be developed here.



Conclusion

In view of the application of the synthetic data as input to turbine load assessment it is obvious that - desides the realistic spectral modelling - the inclusion of coherend events in the generation scheme is essental for relable modelling. However, further research is necessary to obtimize the match of modelled and empirical characteristics,


Learning objectives
- available schemes for data generation require testing for their match to reality.
- modelling of turbulent time series has to be based on realable spectral models
- and advanced scheme for including non gassian characteristics in the synthetic time series
- not negociating the non Gaussian chaarcteristics will lead to an underestiamation of extreeme increments and thus loads